Transcription-Free Fine-Tuning of Speech Separation Models for Noisy and Reverberant Multi-Speaker Automatic Speech Recognition

TL;DR

This paper introduces a transcription-free training approach for speech separation models that improves automatic speech recognition accuracy in noisy, reverberant multi-speaker scenarios by using embedding differences and a modified permutation invariant training.

Contribution

It presents a novel transcription-free joint training method using embedding differences and guided PIT, eliminating the need for reference transcriptions during training.

Findings

6.4% WER improvement over signal-level loss

Enhanced perceptual measures like STOI

Effective in noisy, reverberant multi-speaker environments

Abstract

One solution to automatic speech recognition (ASR) of overlapping speakers is to separate speech and then perform ASR on the separated signals. Commonly, the separator produces artefacts which often degrade ASR performance. Addressing this issue typically requires reference transcriptions to jointly train the separation and ASR networks. This is often not viable for training on real-world in-domain audio where reference transcript information is not always available. This paper proposes a transcription-free method for joint training using only audio signals. The proposed method uses embedding differences of pre-trained ASR encoders as a loss with a proposed modification to permutation invariant training (PIT) called guided PIT (GPIT). The method achieves a 6.4% improvement in word error rate (WER) measures over a signal-level loss and also shows enhancement improvements in perceptual measures such as short-time objective intelligibility (STOI).